Remote-sensing-driven large-area-scale human activity intensity index construction method

Abstract

The invention belongs to the technical field of human activity intensity estimation, and provides a remote sensing-driven large-area scale human activity intensity index construction method, which comprises the following steps of: processing sorted and screened evaluation indexes by adopting a time sequence flat analysis method, a cubic convolution interpolation method and a linear normalization processing method; constructing a human activity intensity index driven by remote sensing; analyzing the multi-scale applicability of the human activity intensity index driven by remote sensing by gradually reducing the spatial resolution of the human activity evaluation index data; according to the method, the problem that the current large-region-scale human activity intensity index is difficult to comprehensively represent the comprehensive influence of human ecology is solved, and the problem that the existing small-region-scale human activity intensity index excessively depends on region parameters, so that cross-scale data standardization is difficult is solved.

Description

Technical Field

[0001] This invention belongs to the field of human activity intensity estimation technology, and relates to remote sensing estimation of human activity intensity, specifically a remote sensing-driven method for constructing a large-scale human activity intensity index. Background Technology

[0002] Quantitative research on human activities generally takes two perspectives: changes in natural systems caused by human activities and the constituent elements of human activities. Changes in natural systems caused by human activities are mainly evaluated using simple indicator methods and comprehensive evaluation methods. Simple indicator methods focus on selecting single indicators closely related to human activities, while comprehensive evaluation methods assess the intensity of human activities by assigning different weights to multiple related indicators (such as road density, population distribution, land use type, and GDP), such as the Human Footprint Index (HFI). Most of these methods rely on statistical data, and the calculation process is relatively straightforward. Some methods are limited by administrative boundaries and cannot effectively reflect the spatial distribution differences of human activities on a large scale; others rely on subjective assessments, such as expert scoring, the analytic hierarchy process (AHP), and visual interpretation, which are difficult to objectively reveal the spatial distribution patterns and temporal variations of human activities.

[0003] Remote sensing technology has driven the quantitative study of human activities. Data on land cover, landscape fragmentation, and net primary productivity can directly and comprehensively reflect the impact of human activities on the natural environment. Methods for quantitatively characterizing human activity intensity based on land cover have been widely applied in practice, such as reclamation index, land development intensity, comprehensive land use index, and human activity intensity on land surface (HAILS). These indices can quickly conduct large-scale studies on human activity intensity and have significant advantages in characterizing human activities reflected by land cover change. However, in densely populated urban areas, due to limitations in data sources and algorithms, the differences between human activities cannot be hierarchically characterized by land cover change, resulting in only a specific value existing within artificial surfaces. Methods for characterizing human activity intensity based on remote sensing-related ecological parameters have been carried out in several small-scale study areas in my country and have effectively analyzed the driving role of human activities on the ecological environment. However, some of the aforementioned studies used a single indicator to characterize human activities. This approach, by selecting only a representation of a specific aspect of human activity, often has significant limitations and fails to comprehensively reflect the combined role of humans in the ecological environment. Other studies relied on comprehensive vegetation indices and ecological parameters, but this method cannot account for the impact of natural vegetation growth and often weakens the representation of the driving force of human activities. Quantitative methods for human activity intensity based on multi-indicator factors using remote sensing technology often suffer from limitations in obtaining certain indicators due to the consideration of typical regional characteristics when constructing indicative factors. These indicators, such as grazing intensity index and irrigation water volume, are difficult to obtain due to regional limitations, and the lack of unified standards hinders large-scale regional studies. All these limiting factors impede the comprehensiveness and long-term monitoring of quantitative data on human activities at the large-scale. Summary of the Invention

[0004] The purpose of this invention is to address the common dilemma faced by existing human activity intensity indices, which often struggle to reconcile large-scale data with high precision. Large-scale data often relies on a single data source, making it difficult to comprehensively represent multidimensional human activity characteristics. On the other hand, high-precision data, due to its high computational complexity and strong regional characteristics of the data source, is difficult to achieve unified cross-regional modeling and cannot meet the needs of multi-scale applications. Based on remote sensing big data, this invention provides a remote sensing-driven method for constructing a large-scale human activity intensity index.

[0005] A method for constructing a large-scale human activity intensity index based on remote sensing includes the following steps:

[0006] S1. Sort out and screen evaluation indicators related to human activities, and process the evaluation indicators using time series anomaly analysis, cubic convolution interpolation and linear normalization methods.

[0007] The human activity evaluation indicators include land cover data, normalized difference vegetation index, aboveground biomass, leaf area index, vegetation cover, surface temperature, food yield, nighttime light, digital elevation model, spatial data of population and economy, point of interest density, road network density, and building volume data.

[0008] S2. Based on human activity evaluation indicators, construct a remote sensing-driven human activity intensity index RS-HAI;

[0009] S3. By gradually reducing the spatial resolution of human activity evaluation index data, the multi-scale applicability of the remote sensing-driven human activity intensity index RS-HAI is analyzed.

[0010] Furthermore, the specific steps of S1 include:

[0011] S11. Select data that can be interpreted and estimated by remote sensing based on relevant evaluation indicators of human activities;

[0012] S12. Process the evaluation index data;

[0013] Furthermore, the specific steps in S12 include:

[0014] S121. The anomaly interpolation method is used to eliminate the interannual natural fluctuations in ecological parameter data. The specific method is as follows:

[0015] The difference between the current year's data and the 5-year average is used to characterize the short-term disturbance of ecological parameters by human activities in that year. The calculation formula is as follows:

[0016] ;

[0017] In the formula: Represents the row and column coordinates of raster cells; The anomaly is the 5-year value; The parameter value for that year; This is the average value of parameters over 5 years.

[0018] S122. A cubic polynomial fitting is performed using 16 neighboring pixels to ensure that the details of the spatiotemporal continuity dataset are preserved to the greatest extent and to unify the spatial resolution; the interpolation kernel function for the cubic polynomial fitting is:

[0019] .

[0020] S123. Use linear normalization technology to linearly map each dataset to the [0,1] interval;

[0021] The calculation formula is as follows:

[0022] ;

[0023] In the formula: The results are linearly normalized for each dataset; For each dataset; , These are the maximum and minimum values ​​of the dataset, respectively.

[0024] Furthermore, the specific methods of S2 include:

[0025] S21. Based on the HAILS index construction theory for human activity intensity on the land surface, the degree of human disturbance and the characteristics of changes in natural attributes on the land surface are divided into four progressive human activity intensity levels i through the secondary land cover type j: unused level (i=1), with a human activity intensity value range of 0-0.25; utilized level (i=2), with a human activity intensity value range of 0.25-0.50; modified level (i=3), with a human activity intensity value range of 0.50-0.75; and developed level (i=4), with a human activity intensity value range of 0.75-1.00.

[0026] Among them, the unused layer (i=1) is mainly natural cover. The natural properties of the cover surface have not been changed and it has not been utilized, but there may still be simple human disturbance to the land cover, such as trampling and viewing activities.

[0027] Utilization level (i=2): Natural or semi-natural land cover, where human activities mainly involve the use of the natural land surface but the attributes of the natural land cover remain unchanged, such as pastureland and economic forests.

[0028] Transformation level (i=3): Artificial vegetation. Human activities mainly manifest as the utilization of natural cover and have changed the original natural attributes, but have not created artificial barriers. These changes can occur between years or within a year, such as paddy fields, dry land, and orchards.

[0029] Development Level (i=4): Artificial Surface. Human activities are mainly manifested in the addition of artificial layers to the natural earth's surface, hindering the exchange of water, nutrients, and air between the upper and lower layers. Examples of human activities represented by land types such as reservoirs, aquaculture ponds, and construction land are examples. The intensity of human activities at each of the above levels and within each level exhibits a progressive relationship. The highest value of the characteristic threshold at each level coincides with the lowest value at the next lower level, thus linking the levels together.

[0030] S22. Based on the four first-level human activity intensity classifications in S21, other evaluation index factors are used to construct the second level within each first level. Among them, normalized vegetation index, aboveground biomass, leaf area index, vegetation cover, surface temperature, and digital elevation model are used to construct the unused and utilized levels. Normalized vegetation index, aboveground biomass, leaf area index, vegetation cover, surface temperature, food yield, digital elevation model, spatial data of population and economy, point of interest density, and road network density are used to construct the renovation level. The development level is constructed using digital elevation model, nighttime lighting, spatial data of population and economy, point of interest density, road network density, and building volume data.

[0031] S23. The number of clusters k obtained based on the optimal clustering scheme i and human activity level parameters The formula for calculating the RS-HAI index is as follows:

[0032] ;

[0033] ;

[0034] In the formula: A represents the initial value for a single pixel; i represents the intensity level of human activity; A i The starting values ​​for level i are unutilized (0), utilized (0.25), modified (0.5), and developed (0.75), respectively; k i The optimal number of clusters for level i; For a single pixel j, the human activity level parameter.

[0035] Further specific methods in S22 include:

[0036] S221. Based on the principal component analysis method, the selected index factors of each first level i of human activity intensity are subjected to dimensionality reduction. First, the principal component variance contribution rate α of the land cover second-level class j of each first level i is calculated. i, j (x, y); secondly, select the first principal component p of type j in level i. i, j (x, y) is used as a comprehensive characterization index; finally, p of each type j within each first level i is calculated. i, j (x, y) mean .

[0037] S222, Mean of type j within level i obtained by dimensionality reduction using principal component analysis. The optimal number of clusters k for type j within each level i is obtained using the K-Means clustering algorithm. iFirst, through an iterative optimization mechanism, based on the principles of minimizing intra-cluster distance and maximizing inter-cluster difference, different numbers of clusters are traversed to generate a series of candidate models, and the sum of squared errors within clusters is used as the criterion for iterative convergence. Second, the silhouette coefficient is used as a clustering quality evaluation index, and the clustering result with the silhouette coefficient closest to 1 is selected to ensure that the selected number of clusters k i To find the optimal clustering scheme that balances intra-class compactness and inter-class discriminability; finally, based on the selected optimal clustering scheme, the mean of type j within level i is obtained by principal component dimensionality reduction. Sort by size, and calculate the human activity level parameter of type j within level i. .

[0038] Furthermore, the specific methods of S3 include:

[0039] S31. Considering the multi-scale characteristics of remote sensing data, by gradually reducing the spatial resolution of the input data, and based on the land cover resolution gradient, from 10 m, 30 m, 100 m, 250 m, 500 m, 1 km, 5 km, and 10 km, the coefficient of variation (CV) and semivariance function parameters of the RS-HAI index are calculated simultaneously at each resolution:

[0040] ;

[0041] ;

[0042] In the formula: This represents the semivariance value when the interval distance is h; The standard deviation of the RS-HAI index at this resolution; The RS-HAI index is the average value at this resolution; The number of sample point pairs with an interval of h; For position RS-HAI index value at the location; To and RS-HAI index value at a distance h.

[0043] S32. By analyzing the abrupt inflection point of CV as resolution decreases, the robustness threshold (the minimum resolution before a significant increase in CV) is determined. At the same time, the spatial structure preservation is evaluated by combining the evolution characteristics of the semivariogram curve morphology (range stability, sill value increase). When the resolution is higher than the threshold, the CV fluctuates smoothly and the semivariogram parameter remains stable, indicating that the RS-HAI index can resist resolution degradation and maintain the original spatial heterogeneity characteristics within this range. Thus, the robustness threshold of RS-HAI as the spatial resolution of the input data is analyzed, and its spatial applicability is evaluated.

[0044] The beneficial effects of this invention are:

[0045] 1. This invention sorts out and selects reasonable and scientific evaluation indicators, and constructs a scientific, standardized and systematic remote sensing-based human activity intensity index, providing a quantitative and spatial dataset for characterizing the degree of human activity at a large regional scale.

[0046] 2. This invention takes the degree of interaction between human activities and the Earth's surface as its starting point, integrates multi-source heterogeneous remote sensing data and geographic-environmental-socio-economic data, and creates a remote sensing-driven human activity intensity index (RS-HAI). This solves the problem that current large-scale human activity intensity indices cannot fully represent the comprehensive impact of human ecology, and overcomes the problem that existing small-scale human activity intensity indices rely too much on regional parameters, leading to difficulties in cross-scale data standardization. Attached Figure Description

[0047] Figure 1 This is a flowchart of the remote sensing-driven human activity intensity index construction process of this invention;

[0048] Figure 2 Table 1 shows the detailed classification characteristics and indicators of secondary land cover types at a large regional scale;

[0049] Figure 3 The index factor is the RS-HAI index level classification index of remote sensing-driven human activity intensity in Table 2. Detailed Implementation

[0050] The present invention will be further described below with reference to the accompanying drawings:

[0051] A method for constructing a regional-scale human activity intensity index based on remote sensing, such as... Figure 1 As shown, it includes the following steps:

[0052] S1. Sort out and screen evaluation indicators related to human activities, and use time series anomaly analysis, cubic convolution interpolation and linear normalization to process the specific indicators.

[0053] The human activity evaluation indicators include land cover data, normalized vegetation index, aboveground biomass, leaf area index, vegetation coverage, surface temperature, food yield, nighttime light, digital elevation model data, spatial data of population and economy, density of points of interest, road network density, and building volume data.

[0054] The specific steps include:

[0055] S11. Select data that can be interpreted and estimated by remote sensing based on relevant evaluation indicators of human activities;

[0056] S12. Process the selected data;

[0057] S121. In order to eliminate the impact of long-term natural fluctuations or periodic changes in ecological parameter datasets (such as NDVI, LAI) on the data and to more clearly highlight the disturbances of short-term human activities or abnormal events on ecological parameters, the anomaly interpolation method is used to eliminate interannual natural fluctuations.

[0058] The difference between the current year's data and the 5-year average is used to characterize the short-term disturbance of ecological parameters by human activities in that year. The calculation formula is as follows:

[0059] ;

[0060] In the formula: Represents the row and column coordinates of raster cells; The anomaly is the 5-year value; The parameter value for that year; This is the average value of parameters over 5 years.

[0061] S122. To address the issue of spatial resolution differences between different datasets, a cubic polynomial fitting is performed using 16 adjacent pixels to ensure that details of the spatiotemporally continuous dataset are preserved to the greatest extent possible, thus unifying the spatial resolution. The interpolation kernel function for the cubic polynomial fitting is:

[0062] .

[0063] S123. To eliminate the influence of different units during multi-source data fusion, reduce the complexity of subsequent model calculations, and accelerate algorithm convergence efficiency, linear normalization is used to linearly map each dataset to the [0,1] interval. The calculation formula is as follows:

[0064] ;

[0065] In the formula: The results are linearly normalized for each dataset; For each dataset; , These are the maximum and minimum values ​​of the dataset, respectively.

[0066] S2. Based on human activity evaluation indicators, construct a remote sensing-driven human activity intensity index;

[0067] The specific steps include:

[0068] S21. Based on the HAILS index construction theory of human activity intensity on land surface, the degree of human disturbance on the land surface and the characteristics of changes in natural attributes are divided into four progressive human activity intensity levels i through the secondary land cover type j, namely: unused level (i=1), used level (i=2), modified level (i=3) and developed level (i=4).

[0069] The land cover is categorized into three levels: Unused Level (i=1): Primarily natural land cover, where the natural properties of the surface layer remain unchanged and unused, but minor human disturbances such as trampling and landscaping may still occur. Utilized Level (i=2): Natural or semi-natural land cover, where human activities mainly involve the use of the natural surface without altering its properties, such as pastureland and economic forests. Modified Level (i=3): Artificial vegetation, where human activities primarily involve the utilization of natural land cover and alteration of its original natural properties, but without creating artificial barriers. These alterations can occur interannually or intrayear, such as paddy fields, dry land, and orchards. Developed Level (i=4): Artificial surfaces, where human activities mainly involve adding artificial layers to the natural surface, hindering the exchange of water, nutrients, and air between the upper and lower layers, such as reservoirs, aquaculture ponds, and construction land. The intensity of human activity exhibits a progressive relationship across all levels and within each level. The highest value of the feature threshold at each level coincides with the lowest value at the next lower level, thus linking the levels together. Detailed classification results are as follows: Figure 2 As shown in Table 1.

[0070] S22. Based on the four first-level human activity intensity classifications in S21, and using other indicator factors, construct the second level within each first level. The indicator factors are as follows: Figure 3 As shown in Table 2, ecological parameters and digital elevation models characterizing vegetation change are used to construct unused and utilized levels. Ecological parameters characterizing vegetation change, grain yield characterizing changes in cultivated land activity, digital elevation models, population and economic data, point-of-interest density, and road network density are used to construct transformation levels. Development levels are constructed using nighttime lighting, digital elevation models, population and economic data, point-of-interest density, road network density, and building volume data.

[0071] The specific methods of S22 include:

[0072] S221. Based on the principal component analysis method, the selected index factors of each first level i of human activity intensity are subjected to dimensionality reduction. First, the principal component variance contribution rate α of the land cover second-level class j of each first level i is calculated. i, j (x, y); then select the first principal component p of type j in level i. i, j (x, y) is used as a comprehensive characterization index; finally, p of each type j within each first level i is calculated. i, j (x, y) mean .

[0073] S222, Mean of type j within level i obtained based on principal component dimensionality reduction The optimal number of clusters k for type j within each level i is obtained using the K-Means clustering algorithm. i First, through an iterative optimization mechanism, based on the principles of minimizing intra-cluster distance and maximizing inter-cluster difference, different numbers of clusters are traversed to generate a series of candidate models. The sum of squared errors and the sum of squared intra-cluster errors are used as the criteria for convergence of the iteration. Second, the silhouette coefficient is used as the clustering quality evaluation index, and the clustering result with the silhouette coefficient closest to 1 is selected to ensure that the selected number of clusters k i To achieve an optimal clustering scheme that balances intra-class compactness and inter-class discriminability, the optimal clustering scheme is determined. Finally, based on the selected optimal clustering scheme, the mean of type j within level i is obtained through principal component dimensionality reduction. Sort by size, and calculate the human activity level parameter of type j within level i. .

[0074] S23. The number of clusters k obtained based on the optimal clustering scheme i and human activity level parameters The formula for calculating the RS-HAI index is as follows:

[0075] ;

[0076] ;

[0077] In the formula: A represents the initial value for a single pixel; i represents the intensity level of human activity; A i The starting values ​​for level i are unutilized (0), utilized (0.25), modified (0.5), and developed (0.75), respectively; k i The optimal number of clusters for level i; For a single pixel j, the human activity level parameter.

[0078] S3. By gradually reducing the spatial resolution of human activity evaluation index data, the multi-scale applicability of the RS-HAI index is analyzed. The specific method is as follows:

[0079] To address the multi-scale characteristics of remote sensing data, the spatial resolution of the input data is progressively reduced. Based on the land cover resolution gradient, the coefficient of variation (CV) and semivariance function parameters of the RS-HAI index are calculated simultaneously at each resolution: 10 m, 30 m, 100 m, 250 m, 500 m, 1 km, 5 km, and 10 km. The calculation formulas are as follows:

[0080] ;

[0081] ;

[0082] In the formula: This represents the semivariance value when the interval distance is h; The standard deviation of the RS-HAI index at this resolution; The RS-HAI index is the average value at this resolution; The number of sample point pairs with an interval of h; For position RS-HAI index value at the location; To and RS-HAI index value at a distance h.

[0083] The robustness threshold, i.e. the minimum resolution before a significant increase in CV, is determined by analyzing the abrupt inflection point of the coefficient of variation (CV) of the RS-HAI index as resolution decreases. At the same time, the spatial structure retention is assessed by combining the morphological evolution characteristics of the semivariogram curve, including range stability and sill value increase.

[0084] When the resolution is higher than the threshold, the coefficient of variation (CV) of the RS-HAI index fluctuates smoothly and the semivariance parameter remains stable. This indicates that within this range, the RS-HAI index can resist resolution degradation and maintain the original spatial heterogeneity characteristics. Thus, we can analyze the robustness threshold of RS-HAI as the spatial resolution of the input data changes and evaluate its spatial applicability.

[0085] This invention identifies and selects 13 reasonable and scientific evaluation indicators, including land cover, normalized difference vegetation index (NDI), aboveground biomass, leaf area index (LAI), vegetation cover, surface temperature, grain yield, nighttime light, digital elevation model (DEM), spatial data of population and economy, point-of-interest (POI) density, road network density, and building volume data. Taking the degree of interaction between human activities and the land surface as the starting point, it integrates multi-source heterogeneous remote sensing data and geographic-environmental-socio-economic data to construct a remote sensing-driven human activity intensity index (RS-HAI). This index, by inputting data at different resolutions, comprehensively characterizes the spatiotemporal distribution patterns of multi-scale human activity features, solving the problems of comprehensiveness in current large-scale human activity intensity indices and the difficulty in standardizing small-scale human activity intensity indices.

Claims

1. A method for constructing a large-scale human activity intensity index based on remote sensing, characterized in that: Includes the following steps: S1. Sort out and screen evaluation indicators related to human activities, and process the evaluation indicators using time series anomaly analysis, cubic convolution interpolation and linear normalization methods. The human activity evaluation indicators include land cover data, normalized difference vegetation index, aboveground biomass, leaf area index, vegetation cover, surface temperature, food yield, nighttime light, digital elevation model, spatial data of population and economy, point of interest density, road network density, and building volume data. S2. Based on human activity evaluation indicators, construct a remote sensing-driven human activity intensity index; S3. By gradually reducing the spatial resolution of human activity evaluation index data, the multi-scale applicability of remote sensing-driven human activity intensity index is analyzed.

2. The method for constructing a large-scale human activity intensity index based on remote sensing as described in claim 1, characterized in that: The specific steps of S1 include: S11. Select data that can be interpreted and estimated by remote sensing based on relevant evaluation indicators of human activities; S12. Process the evaluation index data: S121. Use anomaly interpolation to eliminate interannual natural fluctuations in ecological parameter data; S122. Use 16 adjacent pixels to perform cubic polynomial fitting to unify the spatial resolution; S123. Use the linear normalization calculation formula to linearly map each dataset to the interval [0,1].

3. The method for constructing a large-scale human activity intensity index based on remote sensing as described in claim 1, characterized in that: The specific method of S2 includes: S21. Based on the theory of constructing the intensity index of human activities on the land surface, the degree of human disturbance on the land surface and the characteristics of changes in natural attributes are divided into four progressive first levels of human activity intensity i through the secondary land cover type j, namely: unused level, utilized level, modified level and developed level. S22. Based on the four first-level human activity intensity classifications in S21, other evaluation index factors are used to construct the second level within each first level. Among them, normalized vegetation index, aboveground biomass, leaf area index, vegetation cover, surface temperature, and digital elevation model are used to construct the unused and utilized levels. Normalized vegetation index, aboveground biomass, leaf area index, vegetation cover, surface temperature, food yield, digital elevation model, spatial data of population and economy, point of interest density, and road network density are used to construct the transformation level. The development level is constructed using digital elevation model, nighttime lighting, spatial data of population and economy, point of interest density, road network density, and building volume data. The specific methods of S22 include: S221. Based on principal component analysis, the dimensionality reduction of the index factors selected for each first level i of human activity intensity is performed. S222, Mean of type j within level i obtained by dimensionality reduction using principal component analysis. The optimal number of clusters k for type j within each level i is obtained using the K-Means clustering algorithm. i Based on the selected optimal clustering scheme, combined with the mean of type j within level i obtained by principal component dimensionality reduction. Sort by size, and calculate the human activity level parameter of type j within level i. ; S23. The number of clusters k obtained based on the optimal clustering scheme i and human activity level parameters The formula for calculating the remote sensing-driven human activity intensity index (RS-HAI) is as follows: ； ； In the formula: A represents the initial value for a single pixel; i represents the intensity level of human activity; A i Let i be the starting value for level i, representing unutilized (i=0), utilized (i=0.25), modified (i=0.5), and developed (i=0.75); k i The optimal number of clusters for level i; For a single pixel j, the human activity level parameter.

4. The method for constructing a large-scale human activity intensity index based on remote sensing as described in claim 2, characterized in that: The specific method of S121 is as follows: The difference between the current year's data and the 5-year average is used to characterize the short-term disturbance of ecological parameters by human activities in that year. The calculation formula is as follows: ； In the formula: Represents the row and column coordinates of raster cells; The anomaly is the 5-year value; The parameter value for that year; This is the average value of parameters over 5 years.

5. The method for constructing a large-scale human activity intensity index based on remote sensing as described in claim 2, characterized in that: The interpolation kernel function for the cubic polynomial fitting is: 。 6. The method for constructing a large-scale human activity intensity index based on remote sensing as described in claim 2, characterized in that: The linear normalization calculation formula is as follows: ； In the formula: The results are linearly normalized for each dataset; For each dataset; , These are the maximum and minimum values ​​of the dataset, respectively.

7. The method for constructing a large-scale human activity intensity index based on remote sensing as described in claim 1, characterized in that: The specific method of S3 includes: S31. Based on the resolution gradient of land cover, gradually reduce the spatial resolution of the input data, and simultaneously calculate the coefficient of variation (CV) and semivariance function parameters of the remote sensing-driven human activity intensity index at each resolution. The calculation formulas are as follows: ； ； In the formula: This represents the semivariance value when the interval distance is h; The standard deviation of the remote sensing-driven human activity intensity index at this resolution; This represents the average value of the remote sensing-driven human activity intensity index at this resolution. The number of sample point pairs with an interval of h; For position The intensity index of human activity driven by remote sensing at the location; To and The intensity index of human activity driven by remote sensing at a distance of h; S32. Analyze the abrupt inflection point of the coefficient of variation (CV) as resolution decreases to determine the robustness threshold. At the same time, combine the range stability of the semivariogram curve morphology evolution characteristics and the sill value increase to evaluate the spatial structure preservation. When the resolution is higher than the robustness threshold, the coefficient of variation (CV) fluctuates gently and the semivariogram parameter remains stable, indicating that the remote sensing-driven human activity intensity index in this range can resist resolution degradation and maintain the original spatial heterogeneity characteristics.

Images